Image Quality and Diagnostic Performance of Accelerated Shoulder MRI With Deep Learning-Based Reconstruction

Shoulder MRI using standard multiplanar sequences requires long scan times. Accelerated sequences have tradeoffs in noise and resolution. Deep learning-based reconstruction (DLR) may allow reduced scan time with preserved image quality. The purpose of this study was to compare standard shoulder MRI sequences and accelerated sequences without and with DLR in terms of image quality and diagnostic performance. This retrospective study included 105 patients (45 men, 60 women; mean age, 57.6 ± 10.9 [SD] years) who underwent a total of 110 3-T shoulder MRI examinations. Examinations included standard sequences (scan time, 9 minutes 23 seconds) and accelerated sequences (3 minutes 5 seconds; 67% reduction), both including fast spin-echo sequences in three planes. Standard sequences were reconstructed using the conventional pipeline; accelerated sequences were reconstructed using both the conventional pipeline and a commercially available DLR pipeline. Two radiologists independently assessed three image sets (standard sequence, accelerated sequence without DLR, and accelerated sequence with DLR) for subjective image quality and artifacts using 4-point scales (4 = highest quality) and identified pathologies of the subscapularis tendon, supraspinatus-infraspinatus tendon, long head of the biceps brachii tendon, and glenoid labrum. Interobserver agreement and agreement between image sets for the evaluated pathologies were assessed using weighted kappa statistics. In 27 patients who underwent arthroscopy, diagnostic performance was calculated using arthroscopic findings as a reference standard. Mean subjective image quality scores for readers 1 and 2 were 10.6 ± 1.2 and 10.5 ± 1.4 for the standard sequence, 8.1 ± 1.3 and 7.2 ± 1.1 for the accelerated sequence without DLR, and 10.7 ± 1.2 and 10.5 ± 1.6 for the accelerated sequence with DLR. Mean artifact scores for readers 1 and 2 were 9.3 ± 1.2 and 10.0 ± 1.0 for the standard sequence, 7.3 ± 1.3 and 9.1 ± 0.8 for the accelerated sequence without DLR, and 9.4 ± 1.2 and 9.8 ± 0.8 for the accelerated sequence with DLR. Interobserver agreement ranged from kappa of 0.813-0.951 except for accelerated sequence without DLR for the supraspinatus-infraspinatus tendon (κ = 0.673). Agreement between image sets ranged from kappa of 0.809-0.957 except for reader 1 for supraspinatus-infraspinatus tendon (κ = 0.663-0.700). Accuracy, sensitivity, and specificity for tears of the four structures were not different ( > .05) among image